In my experience it is easier to change a normalized scheme then a non-normalized. The redundancy and integrity issues of non-normalized schemas are really difficult to deal with when the schema has to evolve due to changing business needs. I have certainly never observed that un-normalized data should make it easier to evolve a schema - can you provide an example of when this would be the case?
Regarding performance optimization, I think it is valuable to distinguish between the logical model and the physical model. Any kind of caching is a kind of redundancy, but that is not a problem if the database engine guarantees the integrity of the logical model. Relational database engines provides ways to optimize the physical storage and querying of data without compromising the logical model. For example materialized views can provide the same kind of optimization as a denormalized table in the 5-way join case - but without the integrity issues.
Regarding performance optimization, I think it is valuable to distinguish between the logical model and the physical model. Any kind of caching is a kind of redundancy, but that is not a problem if the database engine guarantees the integrity of the logical model. Relational database engines provides ways to optimize the physical storage and querying of data without compromising the logical model. For example materialized views can provide the same kind of optimization as a denormalized table in the 5-way join case - but without the integrity issues.